Unusual dermoscopic patterns of basal cell carcinoma mimicking melanoma

Abstract Background Basal cell carcinoma can simulate melanoma and specific dermoscopic criteria have not yet been defined in a large cohort. Objective To identify dermoscopic “trump” characteristics for differential diagnosis, identify cluster groups and assess the clinical impact of this study's findings. Methods Retrospective, multicentric comparative study of atypical, non‐facial basal cell carcinoma (≥1 seven‐point checklist criteria) and melanoma (with at least one BCC criteria) at dermoscopy. Observed dermoscopic features were used to develop a proposed score. Lesion clusters were defined with hierarchical analysis. Clinical impact was assessed with a blinded reader study following this study's results. Results A total of 146 basal cell carcinoma and 76 melanoma were included. Atypical vascular pattern was common to most lesions (74.5%). Twelve trump features were included in the proposed score (sensitivity 94.1% and specificity 79.5%). Cluster analysis identified 3 basal cell carcinoma and 3 melanoma clusters. Findings improved overall diagnostic accuracy and confidence (26.8% and 13.8%, respectively; p < 0.001). Conclusions These findings support the notion that atypical vascular pattern should be considered a shared feature of both melanoma and atypical basal cell carcinoma. Our proposed score improves diagnostic accuracy and confidence. Absence of pigmented features was associated with lower diagnostic accuracy and confidence.


| INTRODUC TI ON
Basal cell carcinoma (BCC) is the most common malignant skin cancer and is characterized by higher incidence in fair-skin individuals, slow growth and rare cases of metastasis. 1 The most important risk factor for the development of BCC is exposure to ultraviolet light, as confirmed by its frequent onset in photoexposed areas. 2 In addition to clinical examination, dermoscopy has significantly improved BCC diagnostic accuracy, and criteria have been widely described revealing high sensitivity and specificity. 3,4 However, some BCCs may exhibit equivocal dermoscopic patterns, without typical dermoscopic criteria, simulating other malignant skin tumors, such as malignant melanoma (MM). [5][6][7] Non-facial BCCs simulating MMs have been described in literature with an array of predominant parameters, including atypical network, blue-white veil, irregular dots/globules, irregular streaks and atypical vascular pattern. [8][9][10] Presence of irregular streaks, linear irregular and atypical vessels specifically for lower limb BCCs has been reported. 11,12 A previous study of unselected, consecutive facial and non-facial excised BCCs with histopathological diagnosis was retrospectively evaluated to investigate the variability and significance of dermoscopic features. Dermoscopic features suggestive of melanocytic lesions were observed in 40.6% of BCCs and significantly increased in heavily pigmented BCCs. 10 A large, descriptive study of selected, atypical non-facial BCCs with dermoscopic melanocytic features however is lacking. Differential diagnosis with dermoscopy would improve patient management with correct timing of excision and local treatment.
The current study aims to identify "trump" dermoscopic features in a selected cohort of atypical, non-facial BCCs with melanocytic features (at least 1 feature of the seven-point checklist 13 ) compared to atypical, non-facial MMs with at least 1 feature of BCC, for the development of a score useful for differential diagnosis. Secondary objectives include the identification of BCC and MM subtypes and the assessment of this study's clinical impact with a reader study of diagnostic accuracy and confidence.
(BCC or MM) and indicate a level of diagnostic confidence on a VAS scale (1 = not confident at all, 5 = very confident), considered T0.
The two independent evaluators were shown the score and asked to diagnose each image again (T1) with a level of diagnostic confidence, considering the 12 trump features.

| Statistical analysis
Descriptive statistics and complete case analysis were used for all comparisons between groups. Pearson's χ2 test and Fisher's exact test were used to compare categorical variables in univariate analysis. A hierarchical cluster analysis was performed to identify potential homogeneous subgroups of BCC and MM lesions. All variables were included, and the optimal number (k) of clusters was determined with the Calinski and Harabasz stopping method 15 : The largest pseudo-F value indicates the most distinct clustering. After selecting the optimal number of clusters, the cluster characteristics were analysed with the χ2 test. Dendrograms graphically present the grouping of observations at various levels of (dis)similarity. The association between parameters and outcome was assessed using logistic regression (with backward stepwise process). Multivariate logistic regression (stepwise selection method) was used to identify prognostic factors between groups. p < 0.05 defined variable inclusion into the model and "goodness of fit" was evaluated with Hosmer and Lemeshow test; data were expressed as odds ratio (OR) and 95%

| Study population
A total of 146 BCCs and 76 MMs (222 patients, 444 dermoscopy observations) were included. BCC and MM were comparable for patient sex and lesion location, but the average patient age associated with MM diagnosis was significantly younger, as expected (p = 0.005). Clinical features were significantly different among the groups for all features, with the exception of actinic damage in surrounding skin and crust, see Table S2.
The identification of "trump" dermoscopic features and their influence on final histopathological diagnosis was assessed with a multivariable analysis, outlined in Table S3. In order to assess the clinical applicability of our study, we shared these findings with the blinded evaluators. Overall, diagnostic accuracy and confidence levels improved between T0 and T1 by 27% (68.5% to 86.9%) and 14% (2.9 to 3.3), respectively, see Table S4.
A score, based on regression analysis and including 12 dermoscopic features, was then devised for the differential diagnosis of MM/BCC, see Table 2. We elected to use arbitrary multipliers, rather than exact odds ratios to maximize clinical applicability. The models' sensitivity and specificity were 94.08% and 79.45%, respectively.

| Cluster analysis
A hierarchical cluster analysis of BCC and MM lesions identified 3 clusters for each lesion diagnostic group, see Table 1. Cluster groups were homogeneous for demographic characteristics and lesion location (data not shown).
The BCC "hypo/amelanotic-MM like" cluster included 44 lesions (30.1%), characterized principally by trunk and lower limbs lesion location and hypopigmented BCCs (data not shown). Both BCC and MM dermoscopic non-pigment-related criteria were observed in this cluster, with a predominance of MM criteria. In particular, white-red structureless areas and atypical vascular pattern (95.4%, respectively) were significantly more frequently observed in this cluster (p < 0.001). Superficial (short) fine telangiectasia (80.5%), white streaks (79.3%) and ulceration (73.6%) were also frequently observed.
Pigment-related criteria were reported in a minority of cases, see Table 1, Figure 2.
The BCC "pigmented-BCC-type" cluster, included 33 lesions (22.6%), characterized by pigmented BCCs. Cases mainly displayed dermoscopic BCC specific criteria, such as maple leaf-like areas (72.3%) and multiple blue-grey globules (70.8%). Some MM specific pigment-related criteria were also frequently observed, including irregular dots/globules, regression structures and blue-white veil in over half of the lesions. Non-pigment-related features associated with MM diagnosis were less represented in this cluster compared to other clusters, see Table 1, Figure 3.

| Clinical impact of study findings
The trump features were shared with the blinded, independent evaluators who were asked to evaluate the images again considering the 12 features identified. The second evaluations were then compared to their initial evaluations. Both overall diagnostic accuracy and confidence levels significantly improved (26.8% and 13.8%, respectively; p < 0.001), with the greatest improvement noticed in overall BCC diagnostic accuracy, see Table S4.

| DISCUSS ION
In these selected, non-facial BCC lesions mimicking MM at dermoscopy and MM lesions with BCC-associated dermoscopy features, all features of the seven-point checklist and the BCC-associated features were observed, but at different frequencies.
Despite the age differences between the BCC and MM groups (as was expected 1 ), the groups were homogeneous in sex and lesion location. However, according to clinical features, our evaluation confirms that BCC arises more frequently on phototype I/II and more The analysis performed in this study was able to identify 12 dermoscopic features, including both MM-and BCC-associated features, which proved helpful in differentially diagnosing between these complex lesions. These features are presented in the diagnostic score provided, which reports high diagnostic sensitivity and specificity. Further, we have also proven that the findings of this study significantly impact clinical diagnostic outcome. The blinded evaluators' overall diagnostic accuracy and confidence levels were significantly improved after studying the "trump" dermoscopic features we identified.
For further discrimination of lesions, cluster analysis assisted in identifying BCC and MM lesion subtypes, with the main differences between cluster groups being highlighted by the presence or absence of pigment-related and non-pigment-related features.
According to these subgroups, the most difficult lesions to diagnose with the lowest physician confidence levels at T0 and T1 were the "hypo/amelanotic" BCC and MM lesions.